Honest Headline Feature
I think this is the first thing that could be built for the Index with lowest cost to highest benefit ratio
I recently upgraded my OpenAI account and started to use GPT4. I’m impressed. It’s not as huge a leap over 3.5 as I was fearing but it’s still mightily impressive and I don’t even have access to the visual processing features. As I was playing around with it getting it to do various tasks, I realized one immediate task I could ask it to do was to generate a headline for a news article.
You know how Clickbait Titles like to hide the secret in the article? And then you find yourself digging through something that just wasn’t worth it?
Well, fuck that.
GPT4 to the rescue.
In your browser, whether desktop or mobile, here’s how we change that situation around.
Every hour or so, let’s say, you’re going to run a scrape of the websites for fox, cnn, msnbc, nbc, abc, all the big news orgs. Even the web only ones that are popular like Drudge Report, etc.
For those unfamiliar, a scrape means you sent out a bot to read all their text, gather it up, and deliver it to you in a giant ass file. Well, now you parcel out that file by article and give it to GPT-4. You take the article and say “Write a headline for this that reflects a factual summary of the article. Return it in blah blah blah format so I can pull it out of your response.”
You create a repository of all these articles in a giant table with the old headline, the new headline, and the url. This is crucial. The url is what is going to stop them from fighting back by just changing the headline around. You are always going to be able to scrape the url as fast as they can change it.
Now, when the article renders in your social media app, web browser or wherever, the Index intervenes. For those of you who don’t know there’s some computer language around headlines that makes them special. Our computer code goes out and finds the bit that says <hey, I’m too lazy to look up the format but there’s a headline here> You’ll never believe what Lindsay Lohan did with Ryan Seacrest<end headline> and says, yep this a head line in my tables. And then also there’s a url in there so you can actually click the link and have it do something.
Now, it makes a decision as that code is rendering on your computer and checks: do I have a better headline to put in here?
Yes.
So now when you are just clicking around that headline now reads “Lindsey Lohan and Ryan Seacrest said hello to each other and hugged at an IKEA. Nothing else happened.”
On every single headline you read. You are only paying about $0.06 per query to the GPT4 API and you are only doing that once per article per scrape. So you could potentially build your database for this every day for a couple hundred dollars. Then people could sign up for your service and I can’t stress this enough: You would never see clickbait article headlines again.
I am thinking I might just need to actually start building this thing. With GPT4 I honestly have no excuse on the coding front. Time does seem to be short and while I’ve tried to literally just do the whole thing and manage it for free for a bunch of people I keep getting suspicious looks like “And then money how?” So maybe I just need to start a company, implement, build up a user base, turn the users into Digital Citizens and then maybe Dutch can grow up in a world safe from the AI apocalypse if our data ends up being good enough to build something that aligns to our alignment.
I really need a nap. Spent way too much time at IKEA today. I hate that place a lot. How is everyone else and your attempt to save the world?
Now add the next layer: information cross-checks. News A and News B having contradictory beliefs? Sort it out by checking external information and divergent cultural contexts.